Those who forget the past are doomed to quote it badly. The line everyone attributes to George Santayana actually reads "Those who cannot remember the past are condemned to repeat it," from his 1905 Reason in Common Sense. It concerns historical repetition, not quotation accuracy. But the mangled version illustrates the real subject: when a warning about forgetting history gets repeated carelessly, the warning itself becomes the casualty. Citation drift is measurable, traceable, and, with a few habits, correctable.

Mechanism: how a quotation mutates

A quotation is an information chain, and every retelling is a copy operation with errors. This resembles meme evolution: phrases that are short, morally satisfying, and easy to attribute to a famous name outcompete long, hedged, or obscure originals (Dennett, 1990). The Santayana line is now among the most misquoted in the language, and the mistranscription "doomed to repeat" has essentially outbred the original.

Where does the drift enter? Three failure points dominate. First, paraphrase becomes quotation: a vague memory of the gist gets wrapped in quotation marks. Second, attribution laundering: a theme found in one thinker gets credited to whoever seems most quotable, often Churchill, Einstein, or Twain. Third, compression: online formats abbreviate until the qualifier that made the line true disappears. The Santayana shift from "cannot remember" to "forget" is exactly this kind of compression.

Why does the process accelerate online? Verification costs time, and sharing costs none. A quote forwarded without a source loses its provenance in a single hop. Each screenshot and repost strips the last trace of a citation, and the sentence arrives sounding authoritative simply because it is famous. Meme evolution and library science describe the same dynamic from opposite ends: replication rate beats accuracy whenever checking is optional.

The deeper issue is that these quotes do not just lose accuracy; they change meaning. A warning about forgetting history becomes a smug internet aphorism. The original concern, that losing historical memory leads to repeating mistakes, gets flattened into a meme about misquotation. That irony is the whole point of the keyword. When you forget where a quote came from, you do not just quote it incorrectly; you quote it as if it were always this way. The past disappears.

Quantified context: empty citation slots and long tails

Peer-reviewed data on quotation error is thinner than internet lore suggests. Without inventing figures, we can use only general, well-established patterns.

Academic citation analysis repeatedly shows "empty citation" phenomena: statements placed in quotation marks that lack any source or carry an unverifiable one. The specific percentages vary by field and cannot be cited from memory. Studies in information science document high rates of inaccurate retractions and missing sources on the web, with existing link rot at scale. Exact percentages shift with sample and date. Large quotation anthologies, such as Bartlett's Familiar Quotations, trace many entries to a disputed attribution rather than a verified author.

Claim What can be established What still needs a primary source
Santayana original Published 1905, The Life of Reason Exact first printing wording
"Doomed to repeat" variant Common online paraphrase First documented instance
Churchill attributions Many quotes disputed Item-by-item original sources
Twain/Bacon attributions Frequently unverified Manuscript or letter evidence

Timeline table (general patterns, no fabricated dates):

Era Change Effect on quotation fidelity
Pre-print Oral transmission Heavy drift, no fixed text
Print era Authoritative editions Errors fixed but entrenched
Web 1.0 Copy-paste culture Provenance optional
Social media Screenshots, shrug attribution Meaning loss before source loss
AI retrieval Generated quotes High hallucination risk
Verification tools Full-text search Lowers cost of checking

Edge cases: when attribution fails innocently

Misattribution is not always sloppiness. Some errors are structural.

Translation loss. A quote translated twice can drift so far that the original meaning vanishes, yet attribution remains. Editing convention. Older authors condensed or embellished quotes; their versions became the standard reference. Sarcastic originals. Lines spoken ironically get quoted reverently, a common Twain and Swift failure. Composite quotes. Two different passages fused into one, then attributed to one author. Dispute over priority. Two writers said nearly the same thing; the famous name wins.

Misconceptions: the mistakes that persist

Misconception one: "It is in a quote collection, so it is accurate." Anthologies aggregate earlier errors. The famous Santayana phrasing itself is a good example: the compressed version is so widespread that many people treat it as canonical.

Misconception two: "Attribution is the hard part." Usually the opposite. Getting the wording right is easier than finding a verifiable primary source. A phrase with no source at all should be treated as unverified until proven otherwise.

Misconception three: "Cortisol explains why we quote badly." This is a category error. The stress-hormone language that circulates in popular writing is a metaphor; the problem is information behavior, not endocrinology.

Misconception four: "AI answers solve it." Generation models produce fluent, plausible quotes that have never existed. The RAG (retrieval-augmented generation) approach helps only when the retrieved source is verifiable. Without a citation, generated text is not evidence.

The counterintuitive professional caveat: a perfectly accurate quotation can still be misused. The past is frequently quoted correctly and used as a cudgel. Organizations can verify a founder's words and still abuse them out of context. The deeper warning is that memory of the past matters, not just wording.

Practical actions: a verifiable quote workflow

Here is a concrete workflow, arranged as a checklist.

Classify the phrase. Is it a direct quotation, a paraphrase, or a summary? Apply quotation marks only to direct quotes. Label summary as summary. Search for the exact string in quotation marks, plus an early date. Narrow with year ranges where available. Break the phrase into three to five words to catch variant text. Find a verifiable source. Favor an original edition, an archive scan, or a peer-reviewed collection. Record author and year. Log the source. In a shared doc or note, store the source title, year, and the full sentence around the quote. This is the past component: retain the context so the quote survives transmission. Build a comparison table of candidate texts. Compare the candidate attribution against a second independent source. Flag discrepancies for verification. Check for mutation risk. If the source predates the web, search also for later editions that may have altered the phrasing. Re-check if you share the quote. The habit is simple: if sharing, check again. Treat the quote as an investigation, not a transfer.

A practical "past" system approach: maintain a personal quotations file with two fields, "verbatim" and "provenance." For general readers and writers, this is the professional fix. Provide a one-sentence definition: memory of the past is the contextual evidence attached to a quotation, not the prestige of the name attached to it.

Some organizations will need more formal controls. Their citation policies should require direct source verification, not just name and year. A quote that is sourced correctly is still not permission to stretch it. Source accuracy is not permission to misrepresent.

Limitations: what cannot be concluded

This article describes patterns, not proven rates. Without a defined corpus and dates, "most quotes on social media are wrong" is not a supported claim. The Santayana case shows the mechanism clearly, but it does not give a percentage.

Attribution research also has sampling problems. Much of the link-rot evidence comes from web studies with specific samples; those figures are bounded by their methodology and cannot be generalized to every context. Treat every specific percentage used elsewhere as needing its own source.

Finally, verification solves attribution, not interpretation. A correct quote can still be used badly. The accurate preservation of the past is necessary but not sufficient for understanding it.

This article is general educational material, not a scholarly reference, legal opinion, or professional research finding. It does not claim to quantify quotation error rates or guarantee any outcome. For research-grade or academic work, consult a qualified librarian or look up primary sources directly. If you encounter a disputed attribution with real-world consequences, seek an expert review.